Preliminary Exam Talk
Architectures for Resonance Energy Transfer Optical Computing
pangjun at cs.duke.edu
||Wednesday, April 11, 2012
||1:00pm - 3:00pm
||D344 LSRC, Duke
||Bruce Maggs, Daniel Sorin, Chris Dwyer, Benjamin Lee
The proposal presents a potential alternative to CMOS technology to solve the scaling challenges(e.g. power) that CMOS faces as feature sizes approach tens of nanometers. We propose to use DNA self-assembly to fabricate logic using small optical devices called chromophores. A Chromophore absorbs light at a certain wavelength and emits at a lower energy wavelength. A pair of chromophores can use Resonance Energy Transfer(RET) to communicate with each other in near field. Simple RET logic circuits have been fabricated by patterning chromophores on DNA self-assembled grids.
Building complex computer systems with RET logic, we first investigate the simple filtering provided by single chromophores and one of its potential applications---Network-on-Chip (NoC). We remove ring resonators from current nanophotonic NoC and introduce chromophores and Quantum Dot LED(QD_LED)to provide similar functions. However, these changes enable CNoC to be more energy efficient with smaller area compared to current nanophotonic NoC if proper topologies are chosen. Moreover, we can scale up to a 320X320 optical crossbar with CNoC. Preliminary Lab results demonstrate the feasibility of introducing chromophores into NoC. We also propose three methods to use RET logic to build complex computer systems: 1) Cascade multiple levels of RET logic without a non-linear gain media to provide limited functionality; 2) Design an Optical Logic Element (OLE) as a basic computing unit to support complete logic; 3) Use hidden Markov model or Neural network composed of chromophores to provide some complex functions. A SPICE model of OLE is built to evaluate its performance and simulation results show that OLE has 10X better Power-Delay Product(PDP) than current CMOS process.
Advisor(s): Alvin Lebeck